Paper argues frequent context compaction makes long-context LLMs unnecessary

RustingStrata is a pure-Rust inference engine for Qwen GGUF models with no ggml dependency, compiling CUDA kernels at startup via NVRTC and running the same kernels on Vulkan. It supports Qwen3.5/3.6 from 2B up, and the author says it runs Qwen3.8-Flash-Next faster than the original Strata.

VentureBeat analysis argues AI infrastructure costs are spreading beyond GPUs into memory and per-seat software like Microsoft 365. The piece traces how the spending pressure reaches enterprise budgets across the stack.

Alphabet's DeepMind-spun-out drug discovery unit is negotiating roughly $5 billion in new funds, per Bloomberg sources. The round would value Isomorphic at about $50 billion, up from the at-least-$40 billion valuation reported in early October talks. Financing hasn't closed and terms could change.

Two industry PACs, American Mission and Building the Future, spent $4.8M backing South Carolina Senate candidate Darline Graham, who lost. Anthropic-backed Public First had more success with Democrats, while Guardrails Alliance spent over $1M each opposing two losing candidates.

DevRev uses GPT-6.1 Sol to power its Computer and Operator products, balancing reasoning, speed, and cost for enterprise service.

OpenAI built an autonomous pipeline that found, validated, and generated patches for 948 local privilege-escalation vulnerabilities in the Linux kernel. The post frames the work as cutting down the number of paths to root.

Tenspresso, founded by former Google AI chip engineers, is negotiating a financing round that would value it at $500 million, per people familiar with the matter. The company is roughly three months old.